Circuit board deep blind hole processing control method and control system

Abstract

The embodiment of the application provides a circuit board depth blind hole processing control method and a control system, historical processing monitoring data sets containing multiple processing batches are acquired, multi-dimensional feature fusion is performed on the historical processing monitoring data sets, a processing influence feature set is generated, the processing influence feature set is input into a preset precision optimization model, feature causal analysis is performed through the precision optimization model, and a feature influence result is obtained; based on the feature influence result, a processing parameter adjustment strategy is generated; and adaptive adjustment is performed on path planning parameters and energy input parameters of a current processing process according to the processing parameter adjustment strategy. Through the scheme of the application, the correlation law and the causal relationship of multi-dimensional data can be effectively utilized, when the blind hole depth precision, the hole position precision and the hole wall quality are improved, the influence of a single parameter is considered, and the synergistic effect among parameters is also considered, so that more comprehensive and more reliable processing control effects are realized.

Description

Technical Field

[0001] This invention relates to the field of digital control, and more specifically, to a method and control system for controlling the processing of deep blind vias on circuit boards. Background Technology

[0002] The machining of deep blind vias on printed circuit boards is a critical process in the electronics manufacturing industry. Its machining control technology aims to ensure that the depth accuracy, positional accuracy, and wall quality of the blind vias meet design requirements by adjusting machining parameters. Currently, common machining control methods are typically based on single-dimensional parameter adjustments, such as adjusting the corner radius of the machining path based on experience or setting a fixed energy input value, and then verifying the adjustment effect by sampling and testing the blind via quality. However, this approach, relying solely on empirical adjustments of a single parameter and static quality verification, struggles to capture the dynamic correlation between machining path planning, energy input, and blind via quality. It also fails to clearly define the specific impact of different parameters on quality indicators, resulting in insufficient targeting of adjustment strategies. Consequently, problems such as depth deviation, positional offset, or excessive wall roughness frequently occur during machining, affecting the stability and consistency of blind via quality. Summary of the Invention

[0003] This invention provides a method and control system for controlling the processing of deep blind vias on circuit boards.

[0004] In a first aspect, embodiments of the present invention provide a method for controlling the processing of deep blind vias on circuit boards, comprising: acquiring a historical processing monitoring dataset containing multiple processing batches, wherein the historical processing monitoring dataset contains processing path planning data, energy input records, and three-dimensional detection data of blind vias for each processing batch; performing multi-dimensional feature fusion on the historical processing monitoring dataset to generate a processing influence feature set, wherein the processing influence feature set contains the dynamic correlation between path planning features, energy input features, and blind via quality features; inputting the processing influence feature set into a preset accuracy optimization model, performing feature causal analysis through the accuracy optimization model to obtain feature influence results; generating a processing parameter adjustment strategy based on the feature influence results; and adaptively adjusting the path planning parameters and energy input parameters of the current processing process according to the processing parameter adjustment strategy.

[0005] Secondly, embodiments of the present invention provide a control system, comprising: a memory storing a computer program; and a processor for loading the computer program to implement the circuit board depth blind via processing control method as described above.

[0006] The circuit board depth blind via machining control method provided by this invention acquires a historical machining monitoring dataset containing machining path planning data, energy input records, and blind via 3D detection data. This dataset comprehensively covers multi-dimensional information on pre-machining design parameters, in-machining execution parameters, and post-machining quality results, providing a more complete information foundation for machining control. Through multi-dimensional feature fusion processing, a machining influence feature set is generated, containing the dynamic correlation between path planning features, energy input features, and blind via quality features. This reveals the coupling law between machining parameters and quality results as they change with batch, time, or space, avoiding the limitations of traditional methods that only analyze static and fixed correlations. By performing feature causal analysis through a precision optimization model, the feature influence results are obtained, clarifying the influence strength of path planning features on via position accuracy and the influence of energy input features on depth. Compared to traditional analysis methods that rely solely on statistical correlation, this study more accurately identifies key factors affecting machining quality by exploring the role of depth accuracy and the synergistic influence of both on hole wall quality. Based on the characteristic influence results, a machining parameter adjustment strategy is generated, encompassing path planning correction rules, dynamic energy input adjustment schemes, and real-time monitoring of key points. This strategy enables coordinated parameter adjustment across three levels: pre-machining design, in-machining execution, and in-machining feedback, making the adjustment strategy more targeted and systematic. Ultimately, by adaptively adjusting the path planning and energy input parameters of the current machining process, the study effectively utilizes the correlation patterns and causal relationships of multi-dimensional data. This approach considers both the impact of individual parameters and the synergistic effects between parameters when improving blind hole depth accuracy, hole position accuracy, and hole wall quality, thereby achieving a more comprehensive and reliable machining control effect. Attached Figure Description

[0007] Figure 1 This is a flowchart of a method for controlling the processing of deep blind vias on a circuit board, provided by an embodiment of the present invention.

[0008] Figure 2 This is a schematic diagram of the composition of a control system provided in an embodiment of the present invention. Detailed Implementation

[0009] Please see Figure 1 , Figure 1 A flowchart of a method for controlling the processing of deep blind vias on a circuit board, provided by an embodiment of the present invention, is included. This method can be executed by a control system and includes the following steps:

[0010] Step S100: Obtain a historical processing monitoring dataset containing multiple processing batches. The historical processing monitoring dataset includes processing path planning data, energy input records, and blind hole 3D detection data for each processing batch.

[0011] The historical processing monitoring dataset is a collection of monitoring data from multiple processing batches during previous PCB blind via machining processes. Specifically, the processing path planning data comprises pre-planned processing path data for each batch of PCBs, specifying the movement trajectory and sequence of the machining equipment on the PCB for blind via machining, including blind via positioning coordinates and processing path direction. Energy input records record the energy input from the machining equipment to the PCB during processing, including pulse energy magnitude, duration, and energy density, reflecting the energy supply during machining. Blind via 3D inspection data is obtained after each processing batch is completed, performing 3D inspection on the machined blind vias. This data includes blind via depth, hole position coordinates, and hole wall roughness, used to evaluate machining quality and accuracy.

[0012] Step S200: Perform multi-dimensional feature fusion on the historical processing monitoring dataset to generate a processing impact feature set, which includes the dynamic correlation between path planning features, energy input features, and blind hole quality features.

[0013] Multi-dimensional feature fusion involves comprehensively analyzing and processing different types of data (processing path planning data, energy input records, and blind hole 3D detection data) from multiple dimensions within historical processing monitoring datasets to uncover their potential correlations and characteristics. The processing impact feature set is a feature set generated after fusion processing, integrating the dynamic correlations between path planning features, energy input features, and blind hole quality features. Path planning features are extracted from processing path planning data, reflecting the impact of processing path planning on processing results; energy input features are extracted from energy input records, reflecting the impact of energy input on processing results; and blind hole quality features are extracted from blind hole 3D detection data, describing the processing quality of blind holes. The dynamic correlations between these features reflect the patterns of interaction and influence between processing path planning, energy input, and blind hole quality.

[0014] The process of multi-dimensional feature fusion can employ various methods, such as data mining algorithms and machine learning algorithms. Taking feature engineering methods in machine learning as an example, firstly, different types of data are preprocessed, including data cleaning and normalization, to ensure data quality and consistency. Then, feature extraction techniques are used to extract path planning features, energy input features, and blind via quality features from processing path planning data, energy input records, and blind via 3D detection data, respectively. Finally, association analysis algorithms, such as the Apriori algorithm, are used to mine the association rules between these features, thereby generating a processing influence feature set. For example, in a circuit board processing experiment, by fusing features from multiple sets of historical processing monitoring data, it was found that when the corner curvature in the path planning is large, combined with the set energy input pattern, the wall roughness of the blind via will significantly increase; this association is included in the processing influence feature set.

[0015] In one implementation, step S200 can be specifically implemented as the following steps S210-S240:

[0016] Step S210: Extract the processing path planning data for each processing batch from the historical processing monitoring dataset, perform geometric feature analysis on the processing path planning data to obtain path planning features, which include the initial positioning deviation sequence, path corner curvature distribution and multi-hole spacing uniformity pattern.

[0017] Extracting processing path planning data from historical processing monitoring datasets involves filtering out data relevant to the processing path planning for each processing batch from a large dataset. Geometric feature analysis involves performing geometric analysis and processing on the extracted processing path planning data to uncover hidden geometric features. Path planning features are a set of features obtained through geometric feature analysis that reflect the characteristics of the processing path planning and its potential impact on processing results.

[0018] The initial positioning deviation sequence is the sequence of deviations between the actual initial positioning coordinates of each blind hole and the design coordinates on the design drawing during the machining process. It is obtained by calculating the lateral and longitudinal positioning deviations of each blind hole, reflecting the accuracy of the machining equipment during initial positioning. The path corner curvature distribution is the distribution of the radius of curvature at each corner of the machining path, reflecting the degree of curvature of the machining path and the ease with which the equipment can turn. The multi-hole spacing uniformity pattern is the distribution of the difference between the actual spacing and the design spacing between any two blind holes within the same machining batch, reflecting the uniformity of the blind hole distribution.

[0019] As one implementation method, step S210 involves extracting processing path planning data for each processing batch from the historical processing monitoring dataset, performing geometric feature analysis on the processing path planning data to obtain path planning features, which can be specifically implemented as follows: steps S211-S214:

[0020] Step S211: Calculate the deviation of the initial positioning coordinates in the machining path planning data. Using the design coordinates of the design drawing as the reference, calculate the lateral positioning deviation and longitudinal positioning deviation of each blind hole to obtain the initial positioning deviation sequence.

[0021] When calculating the initial positioning deviation, it is essential to first clarify that the design coordinates on the design drawings serve as a standard reference, specifying the ideal position of each blind hole. The initial positioning coordinates in the machining path planning data are the positioning coordinates when the actual planned machining equipment begins machining each blind hole. By comparing the actual initial positioning coordinates with the design coordinates, the deviation values ​​of each blind hole in the horizontal (e.g., X-axis direction) and vertical (e.g., Y-axis direction) directions are calculated.

[0022] For example, for each blind hole, the lateral positioning deviation is obtained by subtracting the designed lateral coordinate from its actual initial lateral positioning coordinate; the longitudinal positioning deviation is obtained by subtracting the designed longitudinal coordinate from its actual initial longitudinal positioning coordinate. Arranging the lateral and longitudinal positioning deviations of each blind hole sequentially forms the initial positioning deviation sequence. For instance, in a batch of 100 blind holes, the design drawing specifies the design coordinates of the first blind hole as (X... 10 Y 10 ), while the initial positioning coordinates in the actual processing path planning data are (X) 11 Y 11 Then the lateral positioning deviation of the blind hole is ΔX1=X 11 -X 10 The longitudinal positioning deviation is ΔY1=Y 11 -Y 10 The lateral and longitudinal positioning deviations of the remaining 99 blind holes were calculated using the same method, resulting in an initial positioning deviation sequence containing 100 sets of lateral and longitudinal positioning deviation values. This sequence can intuitively reflect the accuracy of the processing equipment during initial positioning.

[0023] Step S212: Calculate the curvature of the path trajectory in the processing path planning data, select equidistant sampling points along the direction of travel of the processing path, calculate the curvature radius of three adjacent points, and statistically analyze the curvature distribution frequency at different positions to obtain the curvature distribution of the path corners.

[0024] The path trajectory is the movement route of the processing equipment specified in the processing path planning data. Curvature calculation is performed on it to understand the degree of curvature. Equidistant sampling points are points selected at set intervals along the direction of travel of the processing path; these points are used for subsequent curvature calculations. The radius of curvature is a physical quantity describing the degree of curvature of a curve. By calculating the radius of curvature of three adjacent points, the curvature of the path at various locations can be obtained. The frequency of curvature distribution at different locations is statistically analyzed; that is, the calculated radius of curvature values ​​are classified and statistically recorded, and the frequency of each radius of curvature value is recorded, thus obtaining the curvature distribution of the path corners.

[0025] When calculating the radius of curvature, the sampling interval ratio can be determined based on the total length of the machining path. Equally spaced sampling points are then divided according to the percentage of the path's length from start to end, ensuring that the number of sampling points for short and long paths is positively correlated with the path length. For example, for a machining path of total length L, if the sampling interval ratio is set to 1%, a sampling point is selected every 0.01L from the path's start point. For each sampling point, its preceding, current, and following sampling points are selected to form a three-point sequence. The radius of curvature of this three-point sequence is then calculated using a mathematical formula. Assuming the coordinates of the three sampling points are (x1, y1), (x2, y2), and (x3, y3), the radius of curvature R can be calculated using the following formula: The curvature radius values ​​are labeled as positive or negative curvature according to the direction of travel along the path, and the distribution frequency of positive and negative curvature is statistically analyzed. Based on the minimum turning radius limit of the processing equipment, the curvature radius values ​​are divided into adaptable and inadaptable curvature intervals, and the percentage of curvature occurrences within each interval is statistically analyzed. Finally, the distribution frequency of positive curvature, the distribution frequency of negative curvature, and the interval percentage data are integrated to obtain the curvature distribution at the path corners. For example, in a circuit board processing path, calculations and statistics show that smaller radii of positive curvature occur more frequently, and some curvature radii fall within the inadaptable curvature interval. This may cause difficulties for the processing equipment in turning at these locations, affecting processing accuracy and efficiency.

[0026] As one implementation method, step S212, which involves calculating the curvature of the path trajectory in the processing path planning data, can be specifically implemented as follows: steps S2121-S2125:

[0027] Step S2121: Determine the sampling interval ratio based on the total length of the processing path, and divide the sampling points into equidistant points according to the percentage of the length from the starting point to the ending point of the path, so that the number of sampling points for short paths and long paths is positively correlated with the path length.

[0028] The total length of the machining path is the total distance traversed by the machining equipment when completing a batch of blind hole machining. The sampling interval ratio is a proportional value determined based on the total length of the machining path, which determines the interval between equidistant sampling points. Equidistant sampling points are divided according to a percentage of the path's length from start to end; that is, the machining path from start to end is divided into sections according to a certain percentage, and each section is a sampling point. Ensuring that the number of sampling points for short and long paths is positively correlated with the path length is to guarantee that sampling points are reasonably distributed along the path, regardless of its length, to accurately reflect the path's curvature.

[0029] Step S2122: For each sampling point, select its previous sampling point, current sampling point and next sampling point to form a three-point sequence, and calculate the radius of curvature of the three-point sequence.

[0030] After determining the equidistant sampling points, for each sampling point, its preceding and following sampling points are selected, forming a three-point sequence. This three-point sequence is used to calculate the radius of curvature because three points can uniquely determine a curve segment, thus allowing the calculation of the curve's radius of curvature, reflecting the degree of curvature of the path at that location.

[0031] The radius of curvature of a three-point sequence can be calculated using mathematical geometry. Assuming the coordinates of the three sampling points are (x1, y1), (x2, y2), and (x3, y3), we can first calculate the radius of the circumcircle of the triangle formed by these three points. This circumcircle radius is the radius of curvature of the three-point sequence. The specific calculation process is as follows:

[0032] First, calculate the side length of the triangle:

[0033] ;

[0034] ;

[0035] ;

[0036] Then, to calculate the area S of the triangle, we can use Heron's formula:

[0037] ;in, ;

[0038] Finally, according to the formula Calculate the radius of curvature R.

[0039] Step S2123: Mark the radius of curvature value as positive curvature or negative curvature according to the direction of travel along the path, and count the distribution frequency of positive curvature and negative curvature respectively.

[0040] Labeling the radius of curvature as positive or negative curvature according to the direction of path travel is to distinguish the curvature of the path in different bending directions. Positive and negative curvature are defined based on the direction of the machining path travel. One direction can be designated as the positive direction. When the path's bending direction is consistent with the positive direction, the corresponding radius of curvature is labeled as positive curvature; when the path's bending direction is opposite to the positive direction, the corresponding radius of curvature is labeled as negative curvature.

[0041] The distribution frequencies of positive and negative curvature are statistically analyzed separately. This involves classifying and statistically analyzing all calculated curvature radius values, recording the frequency of each type of curvature, and calculating their proportion within the total curvature radius values. By statistically analyzing the distribution frequencies of positive and negative curvature, the bending direction distribution of the machining path can be understood.

[0042] Step S2124: Based on the minimum turning radius limit of the processing equipment, divide the curvature radius value into an adaptable curvature range and an unadaptable curvature range, and count the percentage of curvature occurrences in each range.

[0043] The minimum turning radius of a machining equipment is limited to the minimum radius of curvature required for it to turn smoothly during operation. This is determined by the mechanical structure and performance of the equipment. If the radius of curvature of the path is smaller than this minimum turning radius, the equipment may not be able to turn properly, leading to decreased machining accuracy or even equipment damage.

[0044] The curvature radius values ​​are divided into adaptable and inadaptable curvature ranges. Specifically, based on the minimum turning radius limit of the machining equipment, all calculated curvature radius values ​​are categorized into two types. The adaptable curvature range is defined as the range where the curvature radius is greater than or equal to the minimum turning radius of the machining equipment; within this range, the machining equipment can turn smoothly for processing. The inadaptable curvature range is defined as the range where the curvature radius is less than the minimum turning radius of the machining equipment; within this range, the machining equipment may not be able to turn normally.

[0045] The percentage of curvature occurrences within each interval is statistically analyzed. Specifically, the number of curvature radius values ​​occurring in both adaptable and incompatible curvature intervals is recorded, and their proportion within the total curvature radius value is calculated. By statistically analyzing the percentage of curvature occurrences within each interval, the adaptable and incompatible curvature conditions in the machining path can be understood, providing a basis for subsequent adjustments to machining path planning or selection of suitable machining equipment. If the percentage of curvature occurrences in incompatible curvature intervals is too high, the machining path needs to be optimized, or machining equipment with a smaller turning radius capability needs to be replaced.

[0046] Step S2125: Integrate the forward curvature distribution frequency, reverse curvature distribution frequency and interval proportion data to obtain the path corner curvature distribution.

[0047] Integrate the data on the frequency of positive curvature distribution, the frequency of negative curvature distribution, and the proportion of curvature occurrences in the adaptive and inadaptable curvature intervals obtained in the previous steps, and process them together to form a complete description of the curvature distribution of the path corner.

[0048] The curvature distribution at the corners of the machining path reflects the distribution of the radius of curvature at each corner, including the curvature distribution in different bending directions and the distribution of adaptable and inadaptable curvature. By integrating this data, a more comprehensive and accurate understanding of the bending characteristics of the machining path can be obtained, providing a strong basis for analyzing the impact of the machining path on machining accuracy and efficiency. After integrating this data, a detailed curvature distribution at the corners of the path can be obtained, showing that the bending direction of the machining path is mainly positive, and most curvatures are adaptable to the machining equipment, but some inadaptable curvatures still require attention.

[0049] Step S213: Perform uniformity analysis on the hole spacing in the processing path planning data, calculate the difference between the actual spacing and the design spacing of any two blind holes in the same batch, and statistically analyze the distribution of the absolute value of the difference to obtain the hole spacing uniformity pattern.

[0050] The uniformity analysis of the hole spacing in the machining path planning data is performed to assess whether the spacing between blind holes within the same machining batch meets the design requirements and is evenly distributed. The actual spacing is the distance between any two blind holes measured after machining; the design spacing is the ideal distance between any two blind holes specified on the design drawings.

[0051] Calculate the difference between the actual spacing and the designed spacing of any two blind holes within the same batch. That is, for each pair of blind holes, subtract the designed spacing from the actual measured spacing to obtain the difference. Statistically analyze the distribution of the absolute values ​​of the differences. That is, take the absolute value of all calculated differences, then classify and statistically analyze these absolute values, recording the number of times and the frequency of differences occurring in different absolute value ranges.

[0052] By analyzing the distribution of the absolute values ​​of the differences, the uniformity pattern of the via spacing can be determined. If the absolute values ​​of the differences are generally small and the distribution is relatively concentrated, it indicates that the via spacing is relatively uniform and the processing accuracy is high. If the absolute values ​​of the differences are large and the distribution is scattered, it indicates that the via spacing is uneven, which may affect the performance of the circuit board and subsequent assembly. For example, in a circuit board manufacturing project, through uniformity analysis of the via spacing, it was found that the actual spacing between some blind vias differed significantly from the designed spacing. This could lead to difficulties in installing electronic components on the circuit board and affect the normal operation of the entire circuit board.

[0053] Step S214: Integrate the initial positioning deviation sequence, path corner curvature distribution, and aperture spacing uniformity pattern to obtain path planning features.

[0054] By integrating the initial positioning deviation sequence, the path corner curvature distribution, and the porosity uniformity pattern, the three features obtained from the previous calculations and analyses are combined to form a complete path planning feature.

[0055] The initial positioning deviation sequence reflects the accuracy of the processing equipment during initial positioning; the path corner curvature distribution reflects the degree of curvature of the processing path and the ease with which the equipment turns; and the via spacing uniformity pattern reflects the uniformity of blind via spacing within the same batch. Integrating these three characteristics allows for a more comprehensive and accurate description of the characteristics of the processing path planning and its potential impact on the processing results. For example, in a specific circuit board processing path planning, the initial positioning deviation sequence shows that the positioning deviation of some blind vias is relatively large, the path corner curvature distribution indicates the existence of some unsuitable curvatures, and the via spacing uniformity pattern shows that the blind via spacing is uneven in some areas. Integrating this information clearly reveals the problems with the path planning, providing a basis for subsequent adjustments to processing parameters and optimization of the path planning.

[0056] Step S220: Extract the energy input record of each processing batch from the historical processing monitoring dataset, perform time-series feature decomposition on the energy input record to obtain energy input features, which include pulse energy fluctuation curve, duration distribution pattern and energy density spatial distribution.

[0057] Energy input records are extracted from historical processing monitoring datasets, meaning data related to the energy input of each processing batch is sifted from a large amount of previously acquired data. Time-series feature decomposition involves analyzing and processing the extracted energy input records in terms of time series characteristics to uncover hidden time-related features. Energy input features, obtained through time-series feature decomposition, are a set of features that reflect the characteristics of the energy input and its potential impact on processing results.

[0058] The pulse energy fluctuation curve represents the change of pulse energy over time during the processing. It is obtained by dividing the pulse energy values ​​in the energy input record into time intervals and statistically analyzing the maximum, minimum, and average energy values ​​within each window, reflecting the fluctuation of pulse energy during processing. The duration distribution pattern represents the distribution pattern of energy duration when processing blind vias of different depths. This is obtained by statistically analyzing the energy duration corresponding to blind vias of different depths and examining the correlation between duration and target depth, demonstrating the relationship between energy duration and blind via depth. The energy density spatial distribution represents the spatial distribution of energy density on the circuit board surface during processing. It is obtained by mapping the energy density value at each processing location to a two-dimensional coordinate system on the circuit board surface, reflecting the spatial distribution of energy.

[0059] As one implementation method, step S220 can be specifically implemented as the following steps S221-S224:

[0060] Step S221: Divide the pulse energy values ​​in the energy input record into time intervals, taking the processing cycle of a single blind hole as the time interval, and count the maximum, minimum and average values ​​of energy in each window to obtain the pulse energy fluctuation curve.

[0061] The pulse energy values ​​in the energy input record are divided into time intervals. That is, the pulse energy values ​​arranged in chronological order in the energy input record are divided into independent time windows according to the processing cycle of a single blind hole. The processing cycle of a single blind hole is the time taken for the processing equipment to complete the processing of a blind hole from the start of processing.

[0062] The maximum, minimum, and average energy values ​​within each window are statistically analyzed. Specifically, within each time window, the maximum and minimum pulse energy values ​​are identified, and their average value is calculated. By connecting these values ​​in chronological order, a pulse energy fluctuation curve can be obtained. This curve reflects the fluctuation of pulse energy within each blind via machining cycle. A relatively stable pulse energy fluctuation curve indicates a stable pulse energy supply, which is beneficial for ensuring machining quality. However, large fluctuations may affect the accuracy and quality of blind via machining. For example, in a circuit board machining process, dividing and statistically analyzing pulse energy values ​​over time intervals reveals a pulse energy fluctuation curve showing significant fluctuations in pulse energy within certain blind via machining cycles. This could lead to inconsistencies in the depth and diameter of these blind vias, affecting the performance of the circuit board.

[0063] Step S222: Analyze the distribution pattern of the duration of action in the energy input record, count the duration of energy action corresponding to blind holes of different depths, analyze the correlation between duration and target depth, and obtain the distribution pattern of duration of action.

[0064] The distribution pattern of energy input duration is analyzed in the energy input records. This involves analyzing and studying the energy duration during the processing of each blind hole in the records to uncover the distribution patterns. The energy duration for blind holes of different depths is statistically analyzed; that is, the processed blind holes are classified according to depth, and the energy duration for each depth category is statistically analyzed separately.

[0065] Analyzing the relationship between energy application time and target depth involves examining statistically obtained data to identify the intrinsic connection between the energy application time and the target depth of the blind hole. For example, regression analysis can be used to establish a mathematical model of the energy application time and target depth to analyze their correlation. If the energy application time and target depth are positively correlated, it indicates that as the blind hole depth increases, a longer energy application time is needed to ensure processing quality. If there is no clear correlation between the two, further adjustments to processing parameters may be necessary.

[0066] By analyzing the correlation between energy application time and target depth, a distribution pattern of application time was obtained. This pattern can provide a reference for processing blind holes of different depths, allowing for reasonable adjustment of energy application time and improving processing efficiency and quality.

[0067] Step S223: Perform spatial mapping on the energy density in the energy input record. Establish a coordinate system with the circuit board surface as a two-dimensional plane, map the energy density value of each processing position to the corresponding coordinate point, and generate an energy density spatial distribution map.

[0068] Spatial mapping is performed on the energy density in the energy input record, that is, mapping the energy density value of each processing position in the energy input record to a specific position on the circuit board surface. A coordinate system is established with the circuit board surface as a two-dimensional plane. Typically, a reference point on the circuit board can be selected as the origin to establish a two-dimensional Cartesian coordinate system. The energy density value of each processing position is mapped to the corresponding coordinate point, that is, according to the coordinates of the processing position, the corresponding energy density value is marked at the corresponding position in the coordinate system.

[0069] Generating an energy density spatial distribution map involves mapping the energy density values ​​of all processed locations onto a coordinate system, then using interpolation methods to estimate and fill in the energy density values ​​of non-processed locations, forming a continuous energy density distribution image. For example, bilinear interpolation can be used to interpolate between adjacent coordinate points to fill in the estimated energy density values ​​for non-processed locations. Finally, the generated energy density spatial distribution map undergoes edge smoothing to eliminate local outliers caused by equipment measurement inaccuracies, resulting in the final energy density spatial distribution. The energy density spatial distribution map visually displays the distribution of energy density on the circuit board surface. Uneven energy density distribution can lead to inconsistent blind via processing quality at different locations on the circuit board. For example, during circuit board processing, generating the energy density spatial distribution map might reveal an area with excessively high energy density, potentially causing blind vias in that area to have excessively large diameters or increased wall roughness, affecting the circuit board's performance.

[0070] As one implementation method, step S223 can be specifically implemented as the following steps S2231-S2235:

[0071] Step S2231: Establish a two-dimensional Cartesian coordinate system consistent with the coordinate system of the processing equipment, with the reference positioning hole of the circuit board as the origin.

[0072] The reference positioning holes on the circuit board are pre-designed holes for positioning, providing accurate positional information. A two-dimensional Cartesian coordinate system is established with the reference positioning holes as the origin, i.e., the position of the reference positioning holes is taken as the origin (0, 0). Then, the directions and scales of the X and Y axes are determined according to the coordinate system orientation and unit length of the processing equipment. Establishing a two-dimensional Cartesian coordinate system consistent with the coordinate system of the processing equipment ensures that the energy density value can be accurately mapped to the actual position on the circuit board surface, facilitating subsequent analysis and processing.

[0073] Step S2232: Normalize the energy density values ​​in the energy input record. Based on the maximum energy density value in the same batch, convert the energy density values ​​at each location into relative values ​​within a preset range.

[0074] Normalizing the energy density values ​​in the energy input records is to eliminate dimensional differences between energy density values ​​at different locations, making them comparable. The maximum energy density value within the same batch is used as the benchmark; that is, the maximum value among all energy density values ​​in the same processing batch is identified and used as the reference value for normalization.

[0075] The energy density values ​​at each location are converted into relative values ​​within a preset range, typically within the range of [0,1]. The specific normalization method involves dividing the energy density value at each location by the maximum energy density value in the same batch to obtain a relative value. Normalizing the energy density values ​​facilitates subsequent calculations and analysis. When generating a spatial distribution map of energy density, the normalized energy density values ​​provide a more intuitive representation of the relative magnitude and distribution of energy density.

[0076] Step S2233: Based on the design layout of the blind hole, map the machining position of each blind hole to the corresponding coordinate point in the Cartesian coordinate system.

[0077] A blind via design layout diagram is a drawing that specifies the position and size of each blind via when designing a circuit board. Based on the blind via design layout diagram, the processing position of each blind via is mapped to the corresponding coordinate point in a Cartesian coordinate system. That is, based on the coordinate information of the blind via in the design layout diagram, the corresponding position is found in the established two-dimensional Cartesian coordinate system.

[0078] By mapping the machining location of each blind via to a corresponding coordinate point in the Cartesian coordinate system, the energy density value can be linked to the specific location of the blind via, providing accurate location information for the subsequent generation of an energy density spatial distribution map. In a circuit board manufacturing project, mapping the machining locations of blind vias to the Cartesian coordinate system revealed that some blind vias were located in areas of high energy density, which may affect the machining quality of these blind vias.

[0079] Step S2234: Perform bilinear interpolation between adjacent coordinate points to fill in the energy density estimate of non-processing locations and generate a continuous energy density spatial distribution map.

[0080] After mapping the energy density value of each blind hole's machining location to its corresponding coordinate point in the Cartesian coordinate system, these coordinate points are discrete and only represent the energy density value at the machining location. To obtain a continuous spatial distribution of energy density, interpolation needs to be performed between adjacent coordinate points to estimate the energy density value at non-machining locations.

[0081] Bilinear interpolation estimates the energy density of a non-processing location based on the energy density values ​​of its four nearest neighboring coordinate points. Specifically, for a non-processing location, the four nearest neighboring coordinate points are found, and the estimated energy density of that location is calculated using the bilinear interpolation formula based on the energy density values ​​of these four points and the relative position of the point to these four points.

[0082] By performing bilinear interpolation between adjacent coordinate points to fill in the energy density estimates at non-processed locations, a continuous spatial distribution map of energy density can be generated. This map provides a more comprehensive view of the energy density distribution on the circuit board surface, offering a more accurate basis for analyzing the impact of energy density on blind via processing quality.

[0083] Step S2235: Smooth the edges of the energy density spatial distribution map to eliminate local outliers caused by the measurement accuracy of the equipment, and obtain the final energy density spatial distribution.

[0084] Edge smoothing of the energy density spatial distribution map is performed to eliminate local outliers caused by factors such as the measurement accuracy of the equipment. During energy density measurement, due to the accuracy limitations of the measuring equipment or other interference factors, some local abnormal energy density values ​​may appear, which may affect the accurate judgment of the spatial distribution of energy density.

[0085] Edge smoothing can be achieved using various methods, such as Gaussian filtering. By smoothing the energy density spatial distribution map and eliminating local outliers, the final energy density spatial distribution is obtained. This final energy density spatial distribution more accurately reflects the actual distribution of energy density on the circuit board surface, providing a reliable basis for subsequent processing parameter adjustments and quality control.

[0086] Step S224: Integrate the pulse energy fluctuation curve, duration distribution law, and energy density spatial distribution map to obtain energy input characteristics.

[0087] By integrating the pulse energy fluctuation curve, the duration distribution pattern, and the energy density spatial distribution map, the three characteristics obtained above are combined to form a complete energy input characteristic.

[0088] The pulse energy fluctuation curve reflects the temporal fluctuation of pulse energy, the duration distribution pattern demonstrates the correlation between energy duration and blind via depth, and the energy density spatial distribution map shows the spatial distribution of energy density on the circuit board surface. Integrating these three features allows for a more comprehensive and accurate description of the energy input characteristics and their potential impact on processing results. For example, in a circuit board processing project, integrating these three features revealed that areas with large pulse energy fluctuations and uneven energy density spatial distribution generally exhibited poor blind via processing quality. Based on the integrated energy input characteristics, processing parameters can be adjusted in a targeted manner to optimize energy input and improve processing quality and efficiency.

[0089] Step S230: Extract the three-dimensional blind hole detection data of each processing batch from the historical processing monitoring dataset, split the three-dimensional blind hole detection data into quality dimensions, and obtain the blind hole quality features. The blind hole quality features include the depth deviation distribution map, the set of hole position offset trajectories, and the hole wall roughness gradient matrix.

[0090] Blind hole 3D inspection data is extracted from historical processing monitoring datasets, specifically by filtering out data relevant to the 3D inspection of blind holes for each processing batch from a large dataset. Quality dimensions are then broken down into different aspects of the extracted blind hole 3D inspection data to extract features related to blind hole quality. The blind hole quality features, derived from this quality dimension breakdown, reflect the quality and precision of blind hole processing.

[0091] The depth deviation distribution map shows the distribution of deviations between the actual depth and the designed depth of all blind holes within the same processing batch. It is obtained by analyzing the deviation of depth detection values ​​in the 3D blind hole inspection data, using the designed depth as a reference value, calculating the depth deviation value of each blind hole, and statistically analyzing the probability distribution of deviation values ​​for blind holes at different locations, reflecting the consistency and accuracy of blind hole depth. The hole position offset trajectory set represents the offset trajectory between the actual hole position coordinates and the designed hole position coordinates for each blind hole during processing. It is obtained by extracting the offset trajectory of the hole position coordinates from the 3D blind hole inspection data, using the designed hole position coordinates as a reference point, calculating the lateral and longitudinal offsets of the actual hole position, and arranging the offset sequence according to the processing order, reflecting the accuracy of the blind hole positions. The hole wall roughness gradient matrix shows the variation of hole wall roughness along the depth direction of the blind hole. It is obtained by performing gradient analysis of the hole wall roughness in the 3D blind hole inspection data, dividing the hole into equidistant inspection layers along the hole depth direction, statistically analyzing the mean roughness of each layer, and calculating the roughness change rate between adjacent layers, reflecting the uniformity and quality of the hole wall roughness.

[0092] As one implementation method, step S230 can be specifically implemented as the following steps S231-S234:

[0093] Step S231: Perform deviation analysis on the depth detection values ​​in the three-dimensional detection data of blind holes. Using the design depth as the benchmark, calculate the depth deviation value of each blind hole, statistically analyze the distribution probability of the deviation values ​​of blind holes at different locations, and obtain the depth deviation distribution map.

[0094] Deviation analysis is performed on the depth detection values ​​in the 3D inspection data of blind holes. This involves comparing the actual depth detection value of each blind hole with the design depth specified on the design drawings to identify the deviation between them. Using the design depth as the benchmark, for each blind hole, the actual depth detection value is subtracted from the design depth to obtain the depth deviation value of that blind hole.

[0095] The probability distribution of deviation values ​​for blind holes at different locations is statistically analyzed. This involves classifying and statistically analyzing the depth deviation values ​​of all blind holes, and recording the frequency and occurrence of blind holes within different deviation value ranges. For example, the depth deviation values ​​can be divided into several intervals, and the number of blind holes in each interval and their proportion of the total number of blind holes can be counted.

[0096] By statistically analyzing the probability distribution of deviation values ​​for blind vias at different locations, a depth deviation distribution map is obtained. This map visually displays the distribution of blind via depth deviations. If the depth deviation distribution map shows that the deviation values ​​are concentrated within a small range, it indicates good consistency in blind via depth and high processing accuracy. If the deviation values ​​are scattered and over a large range, it indicates poor consistency in blind via depth, and processing parameters may need to be adjusted. For example, in a circuit board processing project, the depth deviation distribution map obtained through depth deviation analysis shows that the depth deviation of some blind vias exceeds the allowable range. This may lead to inconsistent mounting heights of electronic components on the circuit board, affecting the overall performance of the circuit board.

[0097] Step S232: Extract the offset trajectory of the hole position coordinates in the blind hole 3D detection data. Using the designed hole position coordinates as the reference point, calculate the lateral and longitudinal offset of the actual hole position. Arrange the offset sequence according to the processing order to obtain the hole position offset trajectory set.

[0098] Offset trajectory extraction is performed on the hole position coordinates in the 3D inspection data of blind holes, that is, to find the offset between the actual hole position coordinates and the designed hole position coordinates for each blind hole. Using the designed hole position coordinates as the reference point, for each blind hole, the lateral offset is obtained by subtracting the lateral coordinate of the designed hole position from the lateral coordinate of the actual hole position; the longitudinal offset is obtained by subtracting the longitudinal coordinate of the designed hole position from the longitudinal coordinate of the actual hole position.

[0099] The offset sequence is arranged according to the processing order, that is, the lateral and longitudinal offsets of each blind hole are arranged sequentially according to the processing order to form an offset sequence. This sequence reflects the changing trend of the hole position offset of each blind hole during the processing.

[0100] By arranging the offset sequence according to the processing order, a set of hole position offset trajectories is obtained. This set of hole position offset trajectories can help analyze the positioning accuracy and stability of the processing equipment during the processing process. If the offset trajectory set shows that the offset gradually increases or fluctuates significantly, it indicates that there may be a problem with the positioning of the processing equipment, requiring adjustment and calibration.

[0101] Step S233: Perform gradient analysis on the hole wall roughness in the three-dimensional detection data of blind holes, divide the detection layers at equal intervals along the hole depth direction, count the average roughness of each layer, calculate the roughness change rate between adjacent layers, and obtain the hole wall roughness gradient matrix.

[0102] Gradient analysis of hole wall roughness in 3D blind hole inspection data is performed to understand how the hole wall roughness varies along the hole depth direction. Equally spaced inspection layers are divided along the hole depth direction, meaning the blind hole is divided into several equal inspection layers from the hole opening to the bottom, each with the same thickness.

[0103] The mean roughness of each layer is calculated by averaging the measured roughness values ​​of the hole walls within each detection layer. The roughness variation rate between adjacent layers is calculated by dividing the difference between the mean roughness values ​​of two adjacent layers by the layer thickness.

[0104] By statistically analyzing the average roughness of each layer and calculating the rate of roughness change between adjacent layers, a hole wall roughness gradient matrix is ​​obtained. Each row of this matrix represents a detection layer, and each column represents the rate of roughness change between adjacent layers. The hole wall roughness gradient matrix can visually display the trend of hole wall roughness along the hole depth direction. If the rate of roughness change is small, it indicates that the hole wall roughness is relatively uniform and the processing quality is good; if the rate of roughness change is large, it indicates that the hole wall roughness is uneven, which may affect the fitting accuracy between the blind hole and electronic components.

[0105] Step S234: Integrate the depth deviation distribution map, the hole position offset trajectory set, and the hole wall roughness gradient matrix to obtain the blind hole quality characteristics.

[0106] By integrating the depth deviation distribution map, the hole position offset trajectory set, and the hole wall roughness gradient matrix, the three aspects of features obtained above are combined to form a complete blind hole quality feature.

[0107] The depth deviation distribution map reflects the consistency and accuracy of blind via depth, the via position offset trajectory set reflects the accuracy of blind via position and the positioning stability of the processing equipment, and the via wall roughness gradient matrix shows the uniformity and quality of via wall roughness. Integrating these three features allows for a more comprehensive and accurate description of the quality and precision of blind via machining. For example, in a circuit board machining project, by integrating these three features, it was found that the depth deviation distribution map showed that some blind vias had large depth deviations, the via position offset trajectory set indicated that some blind vias had significant position offsets, and the via wall roughness gradient matrix showed that the via wall roughness was uneven. Combining this information, it can be determined that the machining quality of this batch of blind vias is poor, and the machining parameters need to be adjusted.

[0108] Step S240: Establish the temporal correspondence and spatial mapping relationship between path planning features, energy input features and blind hole quality features, and generate a set of processing influence features.

[0109] Establishing the temporal and spatial correspondences between path planning features, energy input features, and blind via quality features aims to identify the intrinsic connections between these three sets of features and clarify their mutual influence in time and space. The temporal correspondence refers to the relationship between path planning features, energy input features, and blind via quality features over time; for example, how path planning features and energy input features affect the quality features of blind vias processed during a certain time period. The spatial mapping refers to the correspondence between these three sets of features in space; for example, how path planning features and energy input features at a certain location on a circuit board affect the quality features of blind vias at that location.

[0110] Generating a processing influence feature set involves establishing temporal and spatial mapping relationships to correlate and integrate path planning features, energy input features, and blind via quality features, forming a comprehensive feature set. This set encompasses the dynamic correlation between path planning, energy input, and blind via quality, providing a more comprehensive reflection of the impact of various factors on blind via processing quality. For example, analyzing temporal relationships reveals that when the corner curvature in the path planning is large and the pulse energy fluctuation in the energy input is also large within a certain time period, the wall roughness of the blind vias processed during that time period increases significantly. Analyzing spatial mapping relationships reveals that when the spacing between vias is uneven and the energy density is too high in a certain area of ​​the circuit board's path planning, the depth deviation of the blind vias in that area is large. By establishing these relationships, the generated processing influence feature set can provide a strong basis for subsequent processing parameter adjustments and quality control.

[0111] Step S300: Input the processing influence feature set into the preset accuracy optimization model, perform feature causal analysis through the accuracy optimization model, and obtain the feature influence results.

[0112] The processing influence feature set is input into a pre-defined accuracy optimization model. Specifically, the previously generated processing influence feature set, which includes the dynamic correlation between path planning features, energy input features, and blind hole quality features, is provided as input data to the pre-set accuracy optimization model. The accuracy optimization model is used to analyze the causal relationships between various features during the processing. It can identify the causal connections between different features by processing and analyzing the input processing influence feature set.

[0113] Feature causal analysis analyzes the strength and manner of the causal influence of path planning features and energy input features on blind hole quality features within the accuracy optimization model. Through feature causal analysis, it becomes clear which path planning features and energy input features have a significant impact on blind hole position accuracy, depth accuracy, and hole wall quality, as well as the specific degree and pattern of these influences. The feature influence results are the detailed results output after performing feature causal analysis on the accuracy optimization model regarding the impact of each feature on blind hole quality. These results include information such as the strength of the influence of path planning features on hole position accuracy, the influence coefficient of energy input features on depth accuracy, and the synergistic influence pattern of path planning features and energy input features on hole wall quality. For example, the feature influence results may show that the path corner curvature distribution in the path planning features has a significant impact on hole position accuracy, the pulse energy fluctuation curve in the energy input features is positively correlated with depth deviation, and certain combinations of path planning features and energy input features have a significant synergistic effect on hole wall roughness. These results can provide important basis for subsequent generation of machining parameter adjustment strategies.

[0114] As one implementation method, the accuracy optimization model includes a feature association module, a causal reasoning module, and a result output module; based on this, step S300 can be specifically implemented as the following steps S310-S350:

[0115] Step S310: Receive the processing influence feature set through the feature association module, synchronize the path planning features with the hole offset trajectory set, and perform time alignment processing on the energy input features and the depth deviation distribution map.

[0116] The feature association module receives the processing influence feature set, meaning the feature association module in the accuracy optimization model obtains the previously generated processing influence feature set. Synchronous alignment of the path planning features and the hole position offset trajectory set is to find the correspondence between the path planning features and the hole position offset. Since the path planning features define the processing path, and the hole position offset trajectory set reflects the offset between the actual hole position and the designed hole position, synchronous alignment clarifies what kind of hole position offset will occur under what kind of path planning. For example, features such as the initial positioning deviation sequence and path corner curvature distribution in the path planning features are mapped one-to-one with the offset of each blind hole in the hole position offset trajectory set according to the processing sequence, revealing their temporal and spatial correspondence.

[0117] Time alignment processing of energy input characteristics and depth deviation distribution maps aims to identify the correspondence between energy input characteristics and blind hole depth deviation. Features such as pulse energy fluctuation curves and duration distribution patterns in the energy input characteristics affect the depth of the blind hole. Time alignment processing maps the changes in energy input characteristics within each time period against the depth deviation distribution map of the blind holes processed within that time period, analyzing the impact of energy input on depth deviation. For example, the pulse energy fluctuation curve in the energy input characteristics is matched chronologically with the depth deviation values ​​in each time period of the depth deviation distribution map to observe the synchronous changes in energy fluctuation and depth deviation.

[0118] By synchronizing the path planning features with the borehole offset trajectory set and performing time alignment processing on the energy input features and depth deviation distribution map, an accurate data foundation can be provided for subsequent causal inference, thereby improving the accuracy of feature causal analysis.

[0119] Step S320: Based on the aligned feature data, the causal reasoning module performs a causal strength calculation operation on the path planning features for hole position accuracy, and determines the probability of the influence of the path corner curvature distribution on the hole position offset through conditional probability analysis.

[0120] Based on the aligned feature data—that is, the path planning features and borehole offset trajectory set aligned by the previous feature association module—the causal inference module performs further analysis. The causal strength calculation operation of the path planning features on borehole accuracy is performed to determine the degree of influence of each sub-feature in the path planning features (such as the initial positioning deviation sequence, path corner curvature distribution, and borehole spacing uniformity pattern) on borehole accuracy (i.e., borehole offset).

[0121] Determining the probability of the influence of path corner curvature distribution on hole position offset using conditional probability analysis is a probabilistic statistical method. Conditional probability analysis calculates the probability of an event occurring under certain given conditions. In this invention, it calculates the probability of hole position offset occurring given the known path corner curvature distribution. For example, the path corner curvature distribution can be divided into different categories according to the size of the curvature radius, the distribution of hole position offset in each category can be statistically analyzed, and the probability that the hole position offset exceeds a certain threshold in a certain curvature radius category can be calculated.

[0122] By using this causal strength calculation and conditional probability analysis, the specific ways and extent in which path planning characteristics affect hole position accuracy can be clearly identified. If the calculation results show that the curvature distribution of path corners has a high probability of affecting hole position offset, it indicates that the curvature of path corners has a significant impact on hole position accuracy, and path planning needs to be adjusted during processing to improve hole position accuracy.

[0123] As one implementation method, step S320 can be specifically implemented as the following steps S321-S326:

[0124] Step S321: Extract the radius of curvature value of the path corner curvature distribution and the corresponding hole offset from the aligned feature data.

[0125] The curvature radius values ​​of the path corner curvature distribution and the corresponding hole position offsets are extracted from the aligned feature data. Specifically, from the path planning features and hole position offset trajectory set aligned by the previous feature association module, the curvature radius values ​​related to the path corner curvature distribution and the hole position offsets of the blind holes corresponding to each curvature radius value are selected. For example, in the aligned feature data, for each path corner, its curvature radius value is recorded, and the hole position offset of the blind hole machined at that corner is found.

[0126] Extracting this data provides a concrete data foundation for subsequent causal strength calculations. This data reflects the direct correspondence between path corner curvature and borehole offset, which helps analyze the impact of path corner curvature on borehole accuracy.

[0127] Step S322: Divide the radius of curvature values ​​into the first curvature group, the second curvature group, and the third curvature group, and calculate the average value of the hole position offset in each group.

[0128] Dividing the radius of curvature values ​​into three groups—High, Medium, and Low—is for the purpose of classifying and analyzing the curvature of path corners. This division can be based on the range of the radius of curvature; for example, radii greater than 10mm are classified into the High group, 5-10mm into the Medium group, and less than 5mm into the Low group. The average hole offset within each group is then calculated, meaning the average of all hole offsets within each of the three groups is taken. For example, for the High group, the average hole offset is obtained by summing all hole offsets and then dividing by the number of holes in that group.

[0129] By grouping the curvature radius values ​​and statistically analyzing the average hole position offset within each group, we can initially observe the influence trend of different curvature radii on hole position offset. If the average hole position offset of the first curvature group is smaller, while the average hole position offset of the third curvature group is larger, it indicates that the smaller the curvature radius, the larger the hole position offset may be, and the curvature of the path corner has a significant impact on hole position accuracy.

[0130] Step S323: Calculate the difference between the average offset of the first curvature group and the average offset of the third curvature group to obtain the basic influence of curvature change on hole position offset.

[0131] Calculate the difference between the mean offset of the first curvature group and the mean offset of the third curvature group, i.e., subtract the mean hole position offset of the third curvature group from the mean hole position offset of the first curvature group. This difference reflects the degree of influence of the change from high curvature radius (first curvature group) to low curvature radius (third curvature group) on the fundamental hole position offset.

[0132] For example, in the previous example, the average hole offset for the first curvature group was 0.1 mm, and the average hole offset for the third curvature group was 0.25 mm. Therefore, the basic influence of curvature change on hole offset is 0.1 - 0.25 = -0.15 mm (the negative sign indicates that the hole offset increases as the radius of curvature decreases). This basic influence can serve as a preliminary indicator to measure the magnitude of the impact of path corner curvature change on hole accuracy. It can help determine the effectiveness of adjusting the path corner curvature radius in improving hole offset during machining. If the basic influence is large, it indicates that adjusting the path corner curvature radius has significant potential to improve hole accuracy.

[0133] Step S324: Introduce the initial positioning deviation sequence as a covariate, and calculate the influence coefficient of the radius of curvature value on the hole position offset through a multiple linear regression model.

[0134] The initial positioning deviation sequence is introduced as a covariate because it may also affect the hole offset. This interference needs to be considered when analyzing the influence of the path corner curvature distribution on the hole offset. Covariates are other variables, besides the independent variable (here, the radius of curvature value), that may affect the dependent variable (here, the hole offset) in the regression analysis.

[0135] The influence coefficient of the radius of curvature on the hole position offset is calculated using a multiple linear regression model. A multiple linear regression model is a statistical model used to analyze the linear relationship between multiple independent variables and a dependent variable. In this invention, the independent variables are the radius of curvature and the initial positioning deviation sequence, and the dependent variable is the hole position offset. The general form of the multiple linear regression model is: Where y is the hole offset, x1 is the radius of curvature, and x2 is the initial positioning deviation sequence. It is the intercept. and It is the regression coefficient. This is the error term.

[0136] The regression coefficients are obtained by fitting a multiple linear regression model using methods such as the least squares method. This coefficient is the influence coefficient of the radius of curvature value on the hole position offset. It reflects the average change in hole position offset for every unit change in the radius of curvature value after considering the influence of the initial positioning deviation sequence.

[0137] Step S325: Use the influence coefficient as the causal strength value of the path corner curvature distribution on the hole position offset.

[0138] The influence coefficient is used as the causal strength value of the path corner curvature distribution on the hole position offset because it directly reflects the degree of influence of the radius of curvature in the path corner curvature distribution on the hole position offset. The larger the causal strength value, the more significant the influence of the path corner curvature distribution on the hole position offset.

[0139] For example, the calculated influence coefficient is -0.05, which can be used as the causal strength value of the path corner curvature distribution on the hole position offset. This negative value indicates that the curvature radius is negatively correlated with the hole position offset; that is, as the curvature radius increases, the hole position offset decreases. By determining the causal strength value, the impact of the path corner curvature distribution in the path planning features on the hole position accuracy can be more accurately assessed, providing a quantitative basis for subsequent adjustments to path planning parameters.

[0140] Step S326: Repeat steps S321-S325 above to calculate the causal strength value of the initial positioning deviation sequence and the hole spacing uniformity pattern on the hole position offset, and obtain the set of hole position influence strength values ​​of the path planning feature.

[0141] For the initial positioning deviation sequence, the values ​​of the initial positioning deviation sequence and the corresponding hole position offsets are extracted from the aligned feature data. The initial positioning deviation sequence is grouped, the mean value of the hole position offset in each group is calculated, the difference between different groups is calculated, other possible covariates (such as the curvature distribution of path corners, etc.) are introduced, and the influence coefficient of the initial positioning deviation sequence on the hole position offset is calculated through a multiple linear regression model. This influence coefficient is used as the causal strength value of the initial positioning deviation sequence on the hole position offset.

[0142] For the pore spacing uniformity mode, a similar operation is performed to determine the causal strength value of the pore spacing uniformity mode on the pore position offset.

[0143] By repeating the above steps, the causal strength values ​​of the initial positioning deviation sequence, path corner curvature distribution, and orifice spacing uniformity pattern on orifice position offset are obtained. Combining these values ​​yields the set of orifice position influence strength values ​​for the path planning features. This set comprehensively reflects the degree of influence of each sub-feature in the path planning features on orifice position accuracy, providing detailed information for evaluating the overall impact of path planning on orifice position accuracy and adjusting path planning parameters.

[0144] Step S330: The causal strength calculation operation of energy input features on depth accuracy is executed synchronously through the causal reasoning module, and the correlation between pulse energy fluctuation curve and depth deviation value is determined through regression analysis.

[0145] The causal inference module synchronously performs the causal strength calculation operation of energy input features on depth accuracy. That is, the causal inference module analyzes the data of aligned energy input features and depth deviation distribution map at the same time to determine the degree of influence of each sub-feature in the energy input features (such as pulse energy fluctuation curve, duration distribution law, energy density spatial distribution, etc.) on depth accuracy (i.e. depth deviation value).

[0146] The correlation between the pulse energy fluctuation curve and the depth deviation value is determined by regression analysis. Regression analysis is a statistical method used to analyze linear or nonlinear relationships between variables. In this invention, relevant parameters in the pulse energy fluctuation curve (such as maximum, minimum, and average energy values) are used as independent variables, and the depth deviation value is used as the dependent variable to establish a regression model and analyze the correlation between the two. For example, a univariate linear regression model can be used, assuming a linear relationship between the depth deviation value y and the average energy value x of the pulse energy fluctuation curve, y = + x+ By fitting this model using methods such as the least squares method, the regression coefficients are obtained. .if The fact that the value is significantly non-zero indicates that there is a linear correlation between the average energy value and the depth deviation value of the pulse energy fluctuation curve.

[0147] By using this causal strength calculation and regression analysis, the specific ways and extent in which energy input characteristics affect depth accuracy can be clearly identified. If the regression analysis results show a positive correlation between the pulse energy fluctuation curve and the depth deviation value, it indicates that fluctuations in pulse energy lead to increased depth deviation, requiring adjustments to the energy input to improve depth accuracy. For example, in a circuit board processing project, regression analysis revealed a positive correlation between the average energy value of the pulse energy fluctuation curve and the depth deviation value; that is, the larger the average energy value, the greater the depth deviation. This suggests that more precise control of the pulse energy is needed to reduce depth deviation.

[0148] As one implementation method, step S330 can be specifically implemented as the following steps S331-S336:

[0149] Step S331: Extract the average energy value and the corresponding depth deviation value of the pulse energy fluctuation curve from the aligned feature data.

[0150] The average energy value and corresponding depth deviation value of the pulse energy fluctuation curve are extracted from the aligned feature data. This is based on the time-aligned data of the energy input features and depth deviation distribution map completed by the previous feature association module. The average energy value of the pulse energy fluctuation curve is the average value of the pulse energy values ​​obtained statistically within each blind hole processing cycle, reflecting the average level of energy input within that cycle. The corresponding depth deviation value is the difference between the actual depth of the blind hole processed in that processing cycle and the designed depth.

[0151] Step S332: Using the average energy value as the independent variable and the depth deviation value as the dependent variable, establish a univariate linear regression model and evaluate the goodness of fit of the model.

[0152] Using the average energy value as the independent variable and the depth deviation value as the dependent variable to establish a univariate linear regression model is a simplified analytical method for preliminarily exploring the impact of the average pulse energy level on depth deviation. The expression for the univariate linear regression model can be found in the aforementioned content or existing techniques, and will not be repeated here.

[0153] The coefficient of determination R is typically used to evaluate the goodness of fit of a model. 2 To measure this. The coefficient of determination, R. 2 The value range of is [0,1], and it represents the proportion of the variation in the dependent variable that can be explained by the independent variable. If R 2 A value close to 1 indicates that the model fits the data well, meaning there is a strong linear relationship between the average energy and the depth deviation; if R... 2 A value close to 0 indicates that the model fits poorly and the energy average has a weak explanatory power for the depth deviation value.

[0154] Step S333: Extract the mean duration of the effect duration distribution pattern from the energy input characteristics, establish a multiple linear regression model with the mean energy and mean duration as the two independent variables and the depth deviation as the dependent variable, and evaluate the explanatory power of the model.

[0155] The mean duration of the energy action is extracted from the energy input characteristics. The distribution pattern of the energy action duration is the distribution of energy action duration corresponding to blind holes of different depths as analyzed earlier. The mean duration is the average value obtained by statistically calculating these action durations, reflecting the average duration of energy action.

[0156] A multiple linear regression model is established with average energy and average duration as the two independent variables and depth deviation as the dependent variable. The general form of the multiple linear regression model can be found in the aforementioned step S324 or existing technology, and will not be elaborated here.

[0157] The coefficient of determination R can also be used to evaluate the explanatory power of a model. 2Indicators such as... Compared to the univariate linear regression model, the multiple linear regression model considers more influencing factors, which may improve its explanatory power for depth bias values.

[0158] Step S334: Compare the fitting effects of univariate regression and multivariate regression to determine the additional impact of duration on depth bias.

[0159] Comparing the fitting effects of univariate and multivariate regression mainly involves comparing the coefficients of determination (R²) of the two models. 2 And other indicators. If the R-squared value of the multiple linear regression model is... 2 The value is significantly greater than the R-squared value of the univariate linear regression model. 2 The value indicates that the model's ability to explain depth deviation values ​​is significantly improved after the mean duration of action is included as an independent variable.

[0160] To determine the additional effect of duration of action on depth deviation, that is, to judge the unique role of duration of action in explaining depth deviation values ​​by comparing the fit results.

[0161] Step S335: Extract the energy density value of the central region of the energy density spatial distribution map from the energy input features, analyze the difference between the blind hole depth deviation value and the edge region corresponding to this region, and obtain the direction of influence of energy density distribution on depth consistency.

[0162] The energy density values ​​of the central region in the energy density spatial distribution map are extracted from the energy input characteristics. This map, generated earlier, reflects the energy density distribution on the circuit board surface, with the central region typically being the middle section. The range of the central region is determined within the energy density spatial distribution map, and the energy density values ​​within that region are extracted. The difference between the blind via depth deviation values ​​in this region and those in the edge regions is analyzed. Specifically, the depth deviation values ​​of blind vias processed in the central and edge regions are statistically analyzed, and their magnitudes and distributions are compared. For example, the average and standard deviation of the blind via depth deviation values ​​in the central and edge regions are calculated, and the differences between them are observed.

[0163] By determining the direction of the influence of energy density distribution on depth consistency, if the energy density value in the central region is higher and the blind hole depth deviation value in this region is generally greater than that in the edge region, then it can be considered that higher energy density leads to increased depth deviation, i.e., energy density distribution has a negative impact on depth consistency. Conversely, if the blind hole depth deviation value in the central region is generally smaller than that in the edge region, it indicates that higher energy density has a positive impact on depth consistency. Through this analysis, the specific spatial impact of energy density distribution on blind hole depth consistency can be clarified, providing a basis for adjusting the spatial distribution of energy input.

[0164] Step S336: Integrate goodness of fit, additional effects, and direction of influence to obtain a set of deep action coefficients for energy input features.

[0165] Integrating goodness of fit, additional effects, and the direction of those effects, this involves considering the goodness of fit (e.g., coefficient of determination R) of the univariate and multivariate linear regression models obtained in the previous steps. 2 ), the additional effect of duration of action on depth deviation (via R) 2 The difference (reflected by the difference) and the influence of energy density distribution on depth consistency are taken into account comprehensively.

[0166] A set of depth impact coefficients for the energy input features is obtained. This set contains information such as the influence coefficients and directions of each sub-feature (pulse energy fluctuation curve, duration distribution, and energy density spatial distribution) on depth accuracy. For example, the set may show the correlation coefficient between the average energy value of the pulse energy fluctuation curve and the depth deviation, the additional influence coefficient of duration on depth deviation, and the direction of influence of energy density distribution on depth deviation in different regions. This set provides a comprehensive understanding of the impact of energy input features on depth accuracy, offering detailed references for subsequent energy input generation adjustment strategies.

[0167] Step S340: Perform the analysis of the synergistic influence of path planning features and energy input features on the hole wall quality through the causal reasoning module, and determine the influence mode of the path corner curvature distribution and energy density spatial distribution on the hole wall roughness gradient matrix through interaction test.

[0168] The reason for performing the analysis of the synergistic influence of path planning features and energy input features on hole wall quality through the causal reasoning module is that in the actual blind via processing of circuit boards, path planning features (such as the curvature distribution of path corners, the uniformity pattern of multi-hole spacing, etc.) and energy input features (such as the spatial distribution of energy density, the pulse energy fluctuation curve, etc.) may interact and jointly affect the hole wall quality (measured by the hole wall roughness gradient matrix).

[0169] This invention uses an interaction test to determine the influence mode of the path corner curvature distribution and energy density spatial distribution on the orifice wall roughness gradient matrix. The interaction test is a statistical analysis method used to determine whether there is an interaction between two or more factors. Specifically, this invention aims to test whether the path corner curvature distribution and energy density spatial distribution jointly affect the orifice wall roughness gradient matrix, and to determine the specific mode of this influence.

[0170] For example, the path corner curvature distribution can be categorized according to the size of the curvature radius, and the energy density spatial distribution can be categorized according to the energy density level. Then, the changes in the orifice wall roughness gradient matrix under different combinations of curvature radius and energy density can be analyzed. If the changes in the orifice wall roughness gradient matrix under certain combinations of curvature radius and energy density are significantly different from other combinations, it indicates an interaction between the path corner curvature distribution and the energy density spatial distribution. Furthermore, the influence pattern of this combination on orifice wall roughness can be summarized, such as whether it increases or decreases orifice wall roughness, and the variation law of this influence in the orifice depth direction.

[0171] As one implementation method, step S340 can be specifically implemented as the following steps S341-S345:

[0172] Step S341: Combine the path corner curvature distribution and aperture spacing uniformity pattern in the path planning features with the energy density spatial distribution and pulse energy fluctuation curve in the energy input features to generate multiple sets of feature combination variables.

[0173] Combining the path corner curvature distribution and pore spacing uniformity patterns from the path planning features with the energy density spatial distribution and pulse energy fluctuation curve from the energy input features is to comprehensively consider the interactions between different features. For example, the path corner curvature distribution can be divided into multiple levels according to the size of the curvature radius (e.g., small curvature radius, medium curvature radius, large curvature radius), the energy density spatial distribution can be divided into multiple regions according to the energy density (e.g., low energy density region, medium energy density region, high energy density region), the pore spacing uniformity pattern can be divided into uniform and non-uniform categories, and the pulse energy fluctuation curve can be divided into different levels according to the fluctuation amplitude.

[0174] Step S342: Use the hole wall roughness gradient matrix in the blind hole quality features as the target variable and input it into the preset random forest regression model.

[0175] The borehole wall roughness gradient matrix from the blind hole quality features is used as the target variable because the aim is to study the influence of path planning features and energy input features on borehole wall quality, and the borehole wall roughness gradient matrix can well reflect the borehole wall quality. A pre-defined random forest regression model is input. The random forest regression model is an ensemble learning model composed of multiple decision trees. In this model, each decision tree analyzes the input feature combination variables and predicts the value of the borehole wall roughness gradient matrix based on different values ​​of the feature combination variables. For example, for the multiple sets of feature combination variables generated earlier, the random forest regression model processes each set of variables and outputs the corresponding predicted value of the borehole wall roughness gradient matrix. By using the random forest regression model, the information in the feature combination variables can be fully utilized to more accurately predict the borehole wall roughness gradient matrix, and the importance of each feature combination variable to the target variable can be evaluated.

[0176] Step S343: Calculate the feature importance score of each feature combination variable to the target variable using a random forest regression model. The feature importance score reflects the explanatory power of the feature combination on the variation of hole wall roughness.

[0177] The random forest regression model calculates the feature importance score of each feature combination variable to the target variable. During training, the random forest regression model calculates the feature importance based on the use of features by each decision tree. For each feature combination variable, the model calculates its importance score in predicting the gradient matrix of hole wall roughness.

[0178] The feature importance score reflects the explanatory power of feature combinations on the variation of hole wall roughness. If the feature importance score of a certain feature combination variable is high, it means that the feature combination has a strong ability to explain the variation of hole wall roughness, that is, the feature combination has a greater influence on the hole wall roughness gradient matrix.

[0179] Step S344: Extract feature combinations whose feature importance scores exceed a preset threshold, and analyze the value range of their corresponding path planning sub-features and energy input sub-features.

[0180] Feature combinations whose importance scores exceed a preset threshold are extracted. This preset threshold is a critical value set based on actual conditions and analytical needs, used to filter out feature combinations that significantly affect changes in pore wall roughness. The value ranges of the corresponding path planning sub-features and energy input sub-features are analyzed. For the extracted feature combinations, the specific value ranges of the path planning sub-features (such as the radius of curvature distribution of path corner curvature, the specific situation of pore spacing uniformity, etc.) and energy input sub-features (such as the energy density value range of energy density spatial distribution, the fluctuation amplitude range of pulse energy fluctuation curve, etc.) are further investigated.

[0181] Step S345: Based on the value range and the target requirements for borehole wall roughness, determine the feature combination identifier of the synergistic influence mode. The feature combination identifier includes the path planning parameters and energy input parameter ranges that need to be controlled simultaneously.

[0182] The values ​​are determined based on the range of values ​​and the target requirements for hole wall roughness. Specifically, the range of values ​​for the path planning sub-features and energy input sub-features that have a significant impact on hole wall quality, obtained from the previous analysis, are considered in conjunction with the target values ​​for hole wall roughness (such as the maximum value of hole wall roughness, uniformity requirements, etc.).

[0183] The feature combination identifier for the synergistic influence mode is determined. This identifier is a label for the corresponding feature combination, which includes the range of path planning parameters (such as the radius of curvature of path corners, aperture spacing, etc.) and energy input parameters (such as energy density, pulse energy, etc.) that need to be controlled simultaneously.

[0184] Step S350: Integrate the causal intensity calculation results and synergistic influence analysis results through the result output module to generate characteristic influence results. The characteristic influence results include the aperture influence intensity value of path planning features, the depth effect coefficient of energy input features, and the characteristic combination identifier of synergistic influence mode.

[0185] The results output module integrates the causal strength calculation results and the synergistic impact analysis results. This module, a component of the accuracy optimization model, summarizes and integrates the various results obtained from the preceding causal reasoning module. The causal strength calculation results include the causal strength values ​​of path planning features on borehole accuracy (such as the influence coefficients of path corner curvature distribution and initial positioning deviation sequence on borehole offset) and the causal strength values ​​of energy input features on depth accuracy (such as the influence coefficients of pulse energy fluctuation curves and duration distribution patterns on depth deviation). The synergistic impact analysis results are obtained from the synergistic impact analysis of path planning features and energy input features on borehole wall quality, such as feature importance scores and feature combination identifiers.

[0186] The feature influence result is a comprehensive set of results, including the hole position influence intensity value of the path planning feature, the depth effect coefficient of the energy input feature, and the feature combination identifier of the synergistic influence mode. This feature influence result provides a comprehensive understanding of the specific impact of path planning features and energy input features on the hole position accuracy, depth accuracy, and hole wall quality of blind holes, offering a complete and accurate basis for subsequent generation of processing parameter adjustment strategies.

[0187] Step S400: Based on the feature impact results, generate a processing parameter adjustment strategy.

[0188] The reason for generating a processing parameter adjustment strategy based on the feature influence results is that the feature influence results obtained above reflect in detail the relationship between path planning features, energy input features and the hole position accuracy, depth accuracy and hole wall quality of blind holes. Based on these relationships, the path planning parameters and energy input parameters in the processing process can be adjusted in a targeted manner.

[0189] For example, if the feature influence results show that the initial positioning deviation sequence in the path planning features has a large influence on the hole position accuracy, it indicates that the accuracy of the initial positioning has a significant impact on the hole position accuracy. Therefore, when generating the processing parameter adjustment strategy, it is necessary to consider how to improve the accuracy of the initial positioning. If the effect coefficient of the pulse energy fluctuation curve in the energy input features shows that the energy fluctuation is positively correlated with the depth deviation, then it is necessary to dynamically adjust the energy input to reduce the impact of energy fluctuation on the depth deviation. If the combination of the path corner curvature distribution and the energy density spatial distribution has a significant impact on the hole wall roughness gradient matrix, then it is necessary to focus on monitoring these key factors during the processing.

[0190] In one implementation, step S400 can be specifically implemented as the following steps S410-S440:

[0191] Step S410: Extract the hole position influence intensity value of the path planning feature from the feature influence results. If the influence intensity value of any type of path planning sub-feature (such as the initial positioning deviation sequence, path corner curvature distribution and multi-hole spacing uniformity mode) exceeds the preset intensity threshold, generate the correction rule for the path planning sub-feature. The correction rule includes the minimum suggested value of the corner radius and the spacing adjustment requirements of the multi-path intersection area.

[0192] Extract the hole position influence intensity value of the path planning feature from the feature influence results, that is, find the influence intensity value of each sub-feature (initial positioning deviation sequence, path corner curvature distribution, multi-hole spacing uniformity pattern, etc.) on the hole position accuracy from the previously generated feature influence results.

[0193] If the influence intensity value of any path planning sub-feature exceeds the preset intensity threshold, the preset intensity threshold is a pre-set critical value used to determine whether the influence of a certain path planning sub-feature on the hole position accuracy is significant.

[0194] The correction rules for generating this path planning sub-feature are as follows: For path planning sub-features that exceed the preset intensity threshold, corresponding correction rules need to be formulated to improve the hole position accuracy. If the influence intensity value of the path corner curvature distribution exceeds the threshold, the correction rules may include a minimum suggested value for the corner radius, such as suggesting that the curvature radius of the path corner should not be less than 5mm, to reduce the hole position offset caused by excessive path curvature. If the influence intensity value of the multi-hole spacing uniformity pattern exceeds the threshold, the correction rules may propose spacing adjustment requirements for multi-path intersection areas, such as requiring that the difference between the blind hole spacing and the design spacing in the multi-path intersection area be controlled within ±0.1mm.

[0195] Step S420: Extract the depth effect coefficient of the energy input feature from the feature influence results. If the effect coefficient of the pulse energy fluctuation curve shows that the energy fluctuation is positively correlated with the depth deviation, then generate a dynamic adjustment scheme for the energy input. The dynamic adjustment scheme includes real-time feedback control logic for pulse energy and energy compensation rules corresponding to different depth targets.

[0196] The depth effect coefficient of energy input features is extracted from the feature influence results. That is, the influence coefficient of each sub-feature (pulse energy fluctuation curve, duration distribution law, energy density spatial distribution, etc.) on depth accuracy is obtained from the feature influence results.

[0197] If the coefficient of the pulse energy fluctuation curve shows a positive correlation between energy fluctuation and depth deviation, it indicates that pulse energy fluctuation leads to an increase in depth deviation. For example, the previous analysis showed that the correlation coefficient between the average energy value of the pulse energy fluctuation curve and the depth deviation is positive and relatively large, indicating that the more intense the energy fluctuation, the greater the depth deviation.

[0198] A dynamic adjustment scheme for generating energy input is needed to reduce the impact of energy fluctuations on depth deviation. This scheme includes real-time feedback control logic for pulse energy, which automatically adjusts the pulse energy magnitude based on the real-time monitoring of pulse energy fluctuations and blind hole depth deviation. For example, when the depth deviation exceeds an allowable threshold, the pulse energy fluctuation amplitude is automatically reduced. Energy compensation rules for different depth targets are formulated for blind holes of different depths, based on their depth targets. For example, for deeper blind holes, the pulse energy input is appropriately increased to ensure the design depth is achieved.

[0199] Step S430: Extract the feature combination identifier of the synergistic influence mode from the feature influence results. If the combination of path corner curvature distribution and energy density spatial distribution has a significant impact on the hole wall roughness gradient matrix, then generate the real-time monitoring focus of the processing process. The real-time monitoring focus includes the energy density monitoring frequency of the corner area and the roughness sampling requirements of the multi-path intersection area.

[0200] Extracting feature combination identifiers of synergistic influence patterns from feature influence results, that is, finding feature combination identifiers of synergistic influence patterns of path planning features and energy input features on borehole wall quality from feature influence results.

[0201] If the combination of path corner curvature distribution and energy density spatial distribution has a significant impact on the borehole wall roughness gradient matrix, it indicates that the combination of these two factors has a significant effect on borehole wall quality. For example, the feature combination of "path corner curvature distribution and energy density spatial distribution" calculated by the random forest regression model has a high feature importance score for the borehole wall roughness gradient matrix, indicating that this combination has a significant impact on borehole wall quality.

[0202] Step S440: Integrate correction rules, dynamic adjustment schemes, and key real-time monitoring points to generate a processing parameter adjustment strategy.

[0203] The integration of correction rules, dynamic adjustment schemes, and real-time monitoring priorities involves comprehensively considering and combining the correction rules for the path planning sub-features generated earlier, the dynamic adjustment schemes for energy input, and the real-time monitoring priorities for the processing process.

[0204] A comprehensive processing parameter adjustment strategy is generated, encompassing methods and monitoring requirements for adjusting path planning and energy input parameters. For example, the strategy might specify adjustments to the curvature radius of path corners and the spacing of multi-path intersections according to correction rules in path planning; real-time adjustment of pulse energy and energy compensation according to a dynamic adjustment scheme in energy input; and real-time monitoring of energy density in corner areas and hole wall roughness in multi-path intersection areas during processing. By generating such a strategy, the processing process can be comprehensively optimized, improving the processing quality of blind vias on circuit boards.

[0205] Step S500: Based on the processing parameter adjustment strategy, adaptively adjust the path planning parameters and energy input parameters of the current processing process.

[0206] Based on the processing parameter adjustment strategy, the path planning parameters and energy input parameters of the current processing process are adaptively adjusted, that is, the previously generated processing parameter adjustment strategy is applied to the actual processing process.

[0207] Regarding path planning parameters, the path planning is adjusted according to the correction rules. For example, if the correction rules suggest a minimum recommended value of 6mm for the radius of curvature of path corners, then in the current processing path planning, the radius of curvature of all path corners is adjusted to be no less than 6mm; if the correction rules require that the difference between the blind hole spacing and the design spacing in the multi-path intersection area be controlled within ±0.1mm, then the blind hole spacing in the multi-path intersection area is checked and adjusted to ensure that the requirements are met.

[0208] Regarding energy input parameters, adjustments are made according to a dynamic adjustment scheme. For example, real-time feedback control logic for pulse energy is activated to monitor pulse energy fluctuations and blind hole depth deviations in real time. When a depth deviation exceeds the allowable threshold, the pulse energy magnitude is automatically adjusted. According to the energy compensation rules corresponding to different depth targets, corresponding energy is input to blind holes of different depths.

[0209] Simultaneously, the processing is monitored in real-time according to key monitoring points. Energy density is monitored at specified frequencies in corner areas of the path; hole wall roughness is randomly inspected in multi-path intersection areas according to sampling requirements. If problems are found during monitoring, further adjustment measures are taken promptly to ensure the processing proceeds according to the adjusted parameters, thereby improving the processing quality and accuracy of blind vias on the circuit board.

[0210] In one implementation, step S500 can be specifically implemented as the following steps S510-S580:

[0211] Step S510: Obtain the design drawings and target blind hole parameters for the current machining task, and extract the corner radius value and multi-path intersection spacing from the design path planning data.

[0212] Obtain the design drawings and target blind via parameters for the current processing task. The design drawings contain detailed design information of the circuit board, such as the location, size, and depth of the blind vias; the target blind via parameters specify the specific requirements for each blind via, such as the design depth and diameter.

[0213] The corner radius and multi-path intersection spacing are extracted from the design path planning data. The design path planning data specifies the movement path of the processing equipment when processing blind holes. The corner radius is the radius of curvature at each corner of the path, which affects the turning difficulty of the processing equipment and the hole position accuracy. The multi-path intersection spacing is the spacing between blind holes on different paths in the multi-path intersection area, which has an important impact on the layout and processing quality of the circuit board.

[0214] Step S520: Adjust the corner radius values ​​in the design path planning data according to the minimum suggested corner radius in the correction rules, so that all corner radii meet the suggested requirements.

[0215] The minimum recommended corner radius in the correction rules refers to the minimum recommended corner radius in the correction rules formulated based on the curvature distribution of path corners in the previously generated processing parameter adjustment strategy. For example, the correction rules recommend that the curvature radius of path corners be no less than 6mm.

[0216] Adjust the corner radius values ​​in the design path planning data. Check each corner radius value recorded in the design path planning data. If any corner radius is found to be smaller than the minimum recommended value, adjust it. For example, if there is a corner radius of 5mm in the design path planning data, which is less than the minimum recommended value of 6mm, adjust the corner radius to 6mm. By making such adjustments, all corner radii will meet the recommended requirements, thereby reducing hole position offset caused by excessive path curvature and improving hole position accuracy.

[0217] Step S530: According to the spacing adjustment requirements of the multi-path intersection area in the correction rules, adjust the blind hole spacing at the multi-path intersection position so that the difference between the actual spacing and the design spacing is within the allowable range.

[0218] According to the spacing adjustment requirements for multi-path intersection areas in the correction rules, the correction rules may specify an allowable range for the difference between the blind via spacing and the design spacing in multi-path intersection areas, such as ±0.1mm. To adjust the blind via spacing at multi-path intersection locations, the blind via spacing in the current processing area is measured and compared with the design spacing. If the difference between the actual spacing and the design spacing exceeds the allowable range, the blind via spacing is adjusted. For example, if the design spacing is 2mm and the measured actual spacing is 2.2mm, exceeding the allowable range of ±0.1mm, then the blind via positions need to be fine-tuned, adjusting the spacing to the range of 1.9-2.1mm. By adjusting the blind via spacing at multi-path intersection locations, the rationality of the circuit board layout and processing quality can be ensured.

[0219] Step S540: Obtain the energy input system parameters of the current processing equipment, and extract the initial set value of pulse energy and the initial set value of pulse duration.

[0220] Obtain the energy input system parameters of the current processing equipment. These parameters include various aspects involved when the processing equipment inputs energy, such as the magnitude, duration, and energy density of the pulse energy. These parameters directly affect the processing quality and accuracy of blind holes.

[0221] Extract the initial set values ​​of pulse energy and the initial set values ​​of the duration of action, that is, find the current set pulse energy magnitude and energy duration from the energy input system parameters.

[0222] Step S550: Based on the real-time feedback control logic in the dynamic adjustment scheme, access the real-time depth monitoring data, and automatically adjust the pulse energy value when the depth deviation exceeds the allowable threshold.

[0223] Based on the real-time feedback control logic in the dynamic adjustment scheme, a logic for adjusting pulse energy in real time according to depth deviation is established. Real-time depth monitoring data is integrated, and during processing, the actual depth data of the blind hole is acquired in real time through depth monitoring equipment. For example, a laser rangefinder or similar device can be used to measure the depth of the blind hole in real time. When a depth deviation exceeding the allowable threshold is detected, the pulse energy value is automatically adjusted. The real-time depth monitoring data is compared with the design depth to calculate the depth deviation. If the depth deviation exceeds the allowable threshold, the pulse energy is automatically adjusted according to the real-time feedback control logic. For example, if the actual depth of the blind hole is detected to be 0.2 mm shallower than the design depth, exceeding the allowable threshold ±0.1 mm, the pulse energy is increased by 1 J according to the real-time feedback control logic to increase the energy input and bring the blind hole to the design depth.

[0224] Step S560: According to the target energy compensation rules for different depths in the dynamic adjustment scheme, increase the basic set value of pulse energy for blind holes at the target depth.

[0225] Based on the energy compensation rules for different depths in the dynamic adjustment scheme, corresponding energy compensation rules are formulated for blind holes of different depths. For example, for a blind hole with a depth of 3mm, 2J of energy needs to be added to the basic pulse energy setting; for a blind hole with a depth of 5mm, 3J of energy needs to be added.

[0226] For blind holes at the target depth, the base setting of the pulse energy is increased. During machining, the base setting of the pulse energy is adjusted according to the target depth of the blind hole and the energy compensation rule. For example, if the target depth of the blind hole being machined is 5mm and the base setting of the pulse energy is 10J, then according to the energy compensation rule, an additional 3J of energy is needed, so the pulse energy is adjusted to 13J. Through this energy compensation, it can be ensured that blind holes of different depths can meet the design requirements, improving depth accuracy.

[0227] Step S570: Based on the energy density monitoring frequency of the corner area in the real-time monitoring focus, set up a high-frequency energy density sensor in the corner area of ​​the processing path and trigger energy density detection at a preset frequency.

[0228] Based on the energy density monitoring frequency in the corner area of ​​the real-time monitoring focus, the real-time monitoring focus specifies the frequency of monitoring energy density in the corner area of ​​the path, such as monitoring once every 3 minutes.

[0229] High-frequency energy density sensors are installed at the corners of the machining path. These sensors can quickly and accurately measure energy density. These sensors are installed at each corner of the machining path to monitor changes in energy density in real time.

[0230] Energy density detection is triggered at a preset frequency. The energy density sensor is periodically activated according to a specified monitoring frequency. For example, every 3 minutes, the sensor automatically collects energy density data from the corner area and transmits the data to the control system. This monitoring allows for the timely detection of abnormal changes in energy density in the corner area, providing a basis for adjusting energy input and ensuring the quality of the borehole wall.

[0231] Step S580: According to the roughness sampling requirements of the multi-path intersection area in the real-time monitoring focus, randomly select blind holes in the multi-path intersection area according to the preset ratio for hole wall roughness detection. If the detection result does not meet the standard, pause processing and trigger parameter readjustment process.

[0232] Based on the roughness sampling requirements for multi-path intersection areas in the real-time monitoring priorities, the real-time monitoring priorities clearly define the proportion and requirements for sampling the roughness of the hole walls in multi-path intersection areas. For example, it requires sampling the roughness of one hole wall every 10 blind holes, and the maximum value of the hole wall roughness cannot exceed 0.5 μm.

[0233] In a multi-path intersection area, blind holes are randomly selected according to a preset ratio for hole wall roughness testing. This can be performed using equipment such as a surface roughness meter. For example, in a multi-path intersection area with 50 blind holes, the 10th, 20th, 30th, 40th, and 50th blind holes are randomly selected for testing, following the requirement of inspecting one blind hole every 10.

[0234] If the test results are substandard, processing is paused and a parameter readjustment process is triggered. If the wall roughness of a blind via exceeds the requirement (e.g., a wall roughness of 0.6 μm, exceeding the maximum value of 0.5 μm), it indicates substandard processing quality. In this case, the current processing is immediately paused, and the parameter readjustment process is triggered. During the readjustment process, the influence of features is re-analyzed, the rationality of the processing parameter adjustment strategy is checked, and further adjustments are made to the path planning parameters and energy input parameters to ensure that the wall roughness of subsequently processed blind vias meets the requirements. This sampling and adjustment mechanism allows for timely detection and resolution of problems during processing, improving the processing quality of blind vias on circuit boards.

[0235] It is understood that the various algorithms, formulas, or models involved in the above descriptions of the embodiments of the present invention can be obtained from relevant content in the prior art. In order to save space, they will not be elaborated or derived in detail in the embodiments of the present invention. In addition, when implementing the solution of the present invention, those skilled in the art can supplement the details based on common knowledge in the art. For example, they can use normalization to eliminate dimensional conflicts before feature fusion, use interpolation to eliminate dimensional differences, reasonably set thresholds based on historical data, experience, or business scenario requirements, train the model based on a general model training method, set the number of layers in the model structure based on actual needs, select activation functions, etc. The present invention will not provide redundant descriptions of overly detailed implementation processes here.

[0236] Please see Figure 2 , Figure 2 This is a schematic diagram of a control system provided in an embodiment of the present invention. The control system is, for example, a computer system installed on a circuit board processing production line, and includes at least a processor 101, a communication interface 102, and a memory 103. The processor 101, communication interface 102, and memory 103 can be connected via a bus or other means. The processor 101 (or Central Processing Unit, CPU) is the computational and control core of the control system, capable of parsing various instructions and processing various data within the control system. The communication interface 102 may optionally include a standard wired interface or a wireless interface (such as Wi-Fi, mobile communication interface, etc.), and can be used to send and receive data under the control of the processor 101; the communication interface 102 can also be used for data transmission and interaction within the control system. The memory 103 is a storage device in the control system used to store programs and data. It is understood that the memory 103 here can include the built-in memory of the control system, or it can include extended memory supported by the control system. The memory 103 provides storage space for storing the operating system of the control system, which may include, but is not limited to, Android, iOS, Windows Phone, etc. The present invention does not limit this.

[0237] In one embodiment, the processor 101 executes the circuit board depth blind via processing control method provided above in the embodiments of the present invention by running a computer program in the memory 103.

Claims

1. A method for controlling the processing of deep blind vias on circuit boards, characterized in that, include: Obtain a historical processing monitoring dataset containing multiple processing batches. The historical processing monitoring dataset includes processing path planning data, energy input records, and blind hole 3D detection data for each processing batch. Multi-dimensional feature fusion is performed on the historical processing monitoring dataset to generate a processing impact feature set, which includes the dynamic correlation between path planning features, energy input features, and blind hole quality features. The processing influence feature set is input into a preset accuracy optimization model, and feature causal analysis is performed through the accuracy optimization model to obtain the feature influence results; Based on the results of the aforementioned feature influence, a processing parameter adjustment strategy is generated. Based on the processing parameter adjustment strategy, the path planning parameters and energy input parameters of the current processing process are adaptively adjusted; The accuracy optimization model includes a feature association module, a causal reasoning module, and a result output module. The process of inputting the processing influence feature set into the preset accuracy optimization model and performing feature causal analysis through the accuracy optimization model to obtain feature influence results includes: The feature association module receives the processing influence feature set, synchronizes the path planning features with the hole offset trajectory set, and performs time alignment processing on the energy input features and the depth deviation distribution map. Based on the aligned feature data, the causal reasoning module performs a causal strength calculation operation on the path planning features for hole position accuracy, and determines the probability of the influence of the path corner curvature distribution on the hole position offset through conditional probability analysis. The causal reasoning module synchronously performs the causal strength calculation operation of energy input characteristics on depth accuracy, and determines the correlation between pulse energy fluctuation curve and depth deviation value through regression analysis. The causal reasoning module performs a synergistic influence analysis of path planning features and energy input features on the hole wall quality. Through interaction testing, the influence mode of the path corner curvature distribution and energy density spatial distribution on the hole wall roughness gradient matrix is ​​determined. The result output module integrates the causal intensity calculation results and the synergistic influence analysis results to generate the characteristic influence results. The characteristic influence results include the aperture influence intensity value of the path planning feature, the depth effect coefficient of the energy input feature, and the feature combination identifier of the synergistic influence mode. The step of performing a causal strength calculation operation on the hole position accuracy based on the aligned feature data through the causal inference module includes: Extract the radius of curvature of the path corner curvature distribution and the corresponding hole offset from the aligned feature data; The radius of curvature values ​​are divided into the first curvature group, the second curvature group, and the third curvature group, and the average value of the hole position offset in each group is calculated. Calculate the difference between the mean offset of the first curvature group and the mean offset of the third curvature group to obtain the basic influence of curvature change on hole position offset; The initial positioning deviation sequence was introduced as a covariate, and the influence coefficient of the radius of curvature on the hole offset was calculated by a multiple linear regression model. The influence coefficient is used as the causal strength value of the path corner curvature distribution on the hole position offset. Repeat the above steps to calculate the causal strength values ​​of the initial positioning deviation sequence and the hole spacing uniformity pattern on the hole position offset, and obtain the set of hole position influence strength values ​​of the path planning features. The step of synchronously performing causal strength calculation operations on depth-accurate energy input features through the causal inference module includes: Extract the average energy value and the corresponding depth deviation value of the pulse energy fluctuation curve from the aligned feature data; A univariate linear regression model was established with the average energy value as the independent variable and the depth deviation value as the dependent variable, and the goodness of fit of the model was evaluated. The mean duration of the effect duration distribution pattern is extracted from the energy input characteristics. The mean energy and mean duration are used as the two independent variables, and the depth deviation is used as the dependent variable. A multiple linear regression model is established to evaluate the explanatory power of the model. Compare the fitting effects of univariate and multivariate regressions to determine the additional impact of duration on depth bias; Extract the energy density value of the central region of the energy density spatial distribution map from the energy input features, analyze the difference between the blind hole depth deviation value and the edge region corresponding to this region, and obtain the direction of the influence of energy density distribution on depth consistency. By integrating the goodness of fit, the additional effects, and the direction of influence, a set of deep action coefficients for the energy input features is obtained.

2. The method as described in claim 1, characterized in that, The process of fusing multi-dimensional features from the historical processing monitoring dataset to generate a processing impact feature set includes: The processing path planning data for each processing batch is extracted from the historical processing monitoring dataset. Geometric feature analysis is performed on the processing path planning data to obtain path planning features. The path planning features include the initial positioning deviation sequence, path corner curvature distribution, and hole spacing uniformity pattern. Energy input records for each processing batch are extracted from the historical processing monitoring dataset. Temporal feature decomposition is performed on the energy input records to obtain energy input features, which include pulse energy fluctuation curves, duration distribution patterns, and spatial distribution of energy density. The blind hole 3D detection data of each processing batch is extracted from the historical processing monitoring dataset. The blind hole 3D detection data is split into quality dimensions to obtain blind hole quality features. The blind hole quality features include depth deviation distribution map, hole position offset trajectory set and hole wall roughness gradient matrix. Establish the temporal correspondence and spatial mapping relationship between the path planning features, the energy input features, and the blind hole quality features, and generate the processing influence feature set.

3. The method as described in claim 2, characterized in that, The process involves extracting processing path planning data for each processing batch from the historical processing monitoring dataset, performing geometric feature analysis on the processing path planning data to obtain path planning features, including: The initial positioning coordinates in the processing path planning data are deviated. Based on the design coordinates of the design drawings, the lateral positioning deviation and longitudinal positioning deviation of each blind hole are calculated to obtain the initial positioning deviation sequence. The curvature of the path trajectory in the processing path planning data is calculated. Equidistant sampling points are selected along the direction of travel of the processing path. The curvature radius of three adjacent points is calculated. The curvature distribution frequency at different positions is statistically analyzed to obtain the curvature distribution of the path corners. The uniformity analysis of the hole spacing in the processing path planning data is performed. The difference between the actual spacing and the design spacing of any two blind holes in the same batch is calculated. The distribution of the absolute value of the difference is statistically analyzed to obtain the hole spacing uniformity pattern. The path planning features are obtained by integrating the initial positioning deviation sequence, the path corner curvature distribution, and the porous spacing uniformity pattern.

4. The method as described in claim 2, characterized in that, The step of extracting energy input records for each processing batch from the historical processing monitoring dataset, and performing time-series feature decomposition on the energy input records to obtain energy input features includes: The pulse energy values ​​in the energy input record are divided into time intervals, with the processing cycle of a single blind hole as the time interval. The maximum, minimum and average values ​​of energy within each window are statistically analyzed to obtain the pulse energy fluctuation curve. The distribution pattern of the duration of action in the energy input record is analyzed, the duration of energy action corresponding to blind holes of different depths is statistically analyzed, and the correlation between duration and target depth is analyzed to obtain the distribution pattern of duration of action. Spatial mapping is performed on the energy density in the energy input record. A coordinate system is established with the circuit board surface as a two-dimensional plane. The energy density value of each processing position is mapped to the corresponding coordinate point to generate an energy density spatial distribution map. By integrating the pulse energy fluctuation curve, the duration distribution law, and the energy density spatial distribution map, the energy input characteristics are obtained.

5. The method as described in claim 2, characterized in that, The process involves extracting blind hole 3D detection data for each processing batch from the historical processing monitoring dataset, and then splitting the blind hole 3D detection data into quality dimensions to obtain blind hole quality features, including: Deviation analysis is performed on the depth detection values ​​in the three-dimensional detection data of the blind holes. Taking the design depth as the benchmark, the depth deviation value of each blind hole is calculated, and the distribution probability of the deviation values ​​of blind holes at different locations is statistically analyzed to obtain a depth deviation distribution map. The offset trajectory is extracted from the hole position coordinates in the blind hole three-dimensional detection data. Taking the designed hole position coordinates as the reference point, the lateral and longitudinal offsets of the actual hole position are calculated. The offset sequence is arranged according to the processing order to obtain the hole position offset trajectory set. Gradient analysis is performed on the hole wall roughness in the three-dimensional detection data of the blind hole. The hole is divided into equally spaced detection layers along the hole depth direction. The roughness mean of each layer is statistically analyzed, and the roughness change rate between adjacent layers is calculated to obtain the hole wall roughness gradient matrix. By integrating the depth deviation distribution map, the hole position offset trajectory set, and the hole wall roughness gradient matrix, the quality characteristics of the blind hole are obtained.

6. The method as described in claim 1, characterized in that, The generation of a processing parameter adjustment strategy based on the influence of the aforementioned features includes: Extract the aperture influence intensity value of the path planning feature from the feature influence results. If the influence intensity value of any type of path planning sub-feature exceeds the preset intensity threshold, generate the correction rule for the path planning sub-feature. The correction rule includes the minimum suggested value of the corner radius and the spacing adjustment requirements of the multi-path intersection area. The depth effect coefficient of energy input features is extracted from the influence results of the features. If the effect coefficient of the pulse energy fluctuation curve shows that the energy fluctuation is positively correlated with the depth deviation, a dynamic adjustment scheme for energy input is generated. The dynamic adjustment scheme includes real-time feedback control logic for pulse energy and energy compensation rules corresponding to different depth targets. Extract the feature combination identifier of the synergistic influence mode from the feature influence results. If the combination of path corner curvature distribution and energy density spatial distribution has a significant impact on the hole wall roughness gradient matrix, then generate the real-time monitoring focus of the processing process. The real-time monitoring focus includes the energy density monitoring frequency of the corner area and the roughness sampling requirements of the multi-path intersection area. The correction rules, the dynamic adjustment scheme, and the real-time monitoring focus are integrated to generate the processing parameter adjustment strategy.

7. The method as described in claim 6, characterized in that, The adaptive adjustment of the path planning parameters and energy input parameters of the current processing process according to the processing parameter adjustment strategy includes: Obtain the design drawings and target blind hole parameters for the current machining task, and extract the corner radius value and multi-path intersection spacing from the design path planning data; Based on the minimum suggested value of the corner radius in the correction rules, adjust the corner radius values ​​in the design path planning data so that all corner radii meet the suggested requirements; According to the multi-path intersection area spacing adjustment requirements in the aforementioned correction rules, adjust the blind hole spacing at the multi-path intersection position so that the difference between the actual spacing and the design spacing is within the allowable range. Obtain the energy input system parameters of the current processing equipment, and extract the initial set value of pulse energy and the initial set value of pulse duration; According to the real-time feedback control logic in the dynamic adjustment scheme, real-time depth monitoring data is accessed, and when the depth deviation is detected to exceed the allowable threshold, the pulse energy value is automatically adjusted. According to the target energy compensation rules for different depths in the dynamic adjustment scheme, the basic set value of pulse energy is increased for blind holes at the target depth; Based on the energy density monitoring frequency of the corner area in the real-time monitoring focus, a high-frequency energy density sensor is set in the corner area of ​​the processing path, and the energy density detection is triggered at a preset frequency. According to the roughness sampling requirements of the multi-path intersection area in the real-time monitoring focus, blind holes are randomly selected in the multi-path intersection area according to a preset ratio for hole wall roughness detection. If the detection result does not meet the standard, the processing is suspended and the parameter readjustment process is triggered.

8. A control system, characterized in that, include: A memory, wherein a computer program is stored; A processor is used to load the computer program to implement the circuit board depth blind via machining control method as described in any one of claims 1-7.

Images